1. Field of the Invention
The present invention relates to blind source separation. More specifically it relates to blind source separation of signals using frequency domain processes.
2. Description of the Related Art
Acoustic mixtures can be separated using blind source separation (BSS) methods, which may comprise either time- or frequency-domain methods. Time-domain methods may achieve better separation performance but may result with more computations than comparable frequency-domain methods. Also, since in time-domain methods, every filter tap adaptation depends on all other taps, convergence may be slow, prone to local minima and may therefore heavily depend on good initialization.
Therefore, there is a need for devices and methods that can result in improved signal separation while reducing computation time and/or improving solution convergence.